In-network slice autonomous anomaly detection and mitigation

Modern mobile networks serve many different users, devices and applications. A video call, for example, may need a different level of speed and responsiveness from an industrial service. Network slicing allows several virtual networks, each designed for different needs, to share the same physical infrastructure.
Keeping these services reliable is difficult. People move between locations, demand changes throughout the day, and different services compete for limited computing and network resources. As a result, connections may slow down or fail to meet the promised level of service. Traditional management methods often respond only after a problem has occurred.

This dissertation explores how artificial intelligence can help mobile networks anticipate problems and manage themselves more effectively. It develops methods for three related tasks: detecting service problems before they happen, reducing their impact, and preventing them where possible.

First, the proposed system examines not only the performance of individual network devices but also how different parts of the network are connected. This broader view helps it identify early signs that service quality may deteriorate. Second, by predicting how users move, the system can decide when to move services between computing locations or adjust the resources assigned to them. Third, it distributes traffic across multiple routes while accounting for different types of requests, helping avoid overloaded parts of the network.

The methods were evaluated through simulations and compared with existing approaches. The results show improvements in early problem detection, resource use, service reliability and user experience, while helping reduce operating costs. Overall, the research contributes to mobile networks that can adapt to changing conditions with less need for manual intervention.